Tethered Monte-Carlo Sampling for Restricted Boltzmann Machines

Using the Tethered Monte-Carlo Sampling is an efficient method to train a Restricted Boltzmann Machine on a low-dimensional dataset. It can also be used to recover the free-energy profile of the RBM and improve the generation of new decorrelated samples on slow PCD trained models.

This repository contains the material used to obtain the results in our paper

Prerequisites

Install python 3:

sudo apt update
sudo apt install python3
sudo apt install python3-pip

Install torch 1.11.0

Install requirements:

pip install -r requirements.txt

Installation

Install the package:

pip install -e .

Usage

You can import the package and use the functions in the rbm module. For example:

from rbm.models import RBM

An example of how to train a RBM using the tethered monte-carlo sampling method is shown in those notebooks :

• Dataset 1D 2 clusters
• Dataset 2D 3 clusters

And a sampling and analysis example is shown in those notebooks :

• Dataset 1D 2 clusters
• Dataset 2D 3 clusters